You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 
mindspore-ci-bot 9edd57eef2 !1268 Bugfix: online inference get const need from context first 3 years ago
..
worker !1224 add process for some op need infershape when running 3 years ago
hybrid_execution_context.cc dynamic shape support pipeline 3 years ago
hybrid_execution_context.h fixed sc warning 3 years ago
hybrid_model_async_executor.cc !1268 Bugfix: online inference get const need from context first 3 years ago
hybrid_model_async_executor.h dynamic shape over flow 3 years ago
hybrid_model_executor.cc infer dump 3 years ago
hybrid_model_executor.h dynamic shape support pipeline 3 years ago
hybrid_model_pipeline_executor.cc infer dump 3 years ago
hybrid_model_pipeline_executor.h fixed sc warning 3 years ago
hybrid_profiler.cc dynamic shape support pipeline 3 years ago
hybrid_profiler.h inference supports dynamic shape 3 years ago
node_done_manager.cc cpplint magic num & define macro 3 years ago
node_done_manager.h update cmake 4 years ago
node_state.cc remove check shape by shape range 3 years ago
node_state.h inference dynamic input 3 years ago
rt_callback_manager.cc dynamic shape support pipeline 3 years ago
rt_callback_manager.h dynamic shape support pipeline 3 years ago
subgraph_context.cc update 3 years ago
subgraph_context.h update 3 years ago
subgraph_executor.cc modified: ge/hybrid/executor/hybrid_model_async_executor.cc 3 years ago
subgraph_executor.h dynamic shape support pipeline 3 years ago

图引擎模块(GE)是MindSpore的一个子模块,其代码由C++实现,位于前端模块ME和底层硬件之间,起到承接作用。图引擎模块以ME下发的图作为输入,然后进行一系列的深度图优化操作,最后输出一张可以在底层硬件上高效运行的图。GE针对昇腾AI处理器的硬件结构特点,做了特定的优化工作,以此来充分发挥出昇腾AI处理器的强大算力。在进行模型训练/推理时,GE会被自动调用而用户并不感知。GE主要由GE API和GE Core两部分组成,详细的架构图如下所示

C++ Text Makefile Shell CMake other